Addressing confabulation

[daveshapautomator]

Preventing ‘Hallucination’ in GPT-3 and Other Complex Language Models

[doctorew]

Machines are totally unable to suffer and to hallucinate!

“AI is neither Artificial nor Intelligent”

[newcolor]

My ex used to do this all the time The ‘Q & A’ preset (davinci): temperature ‘0’ produced pretty deterministic results when I quizzed it on authors. I’m impressed by the APIs ability to shapeshift from deterministic answer-machine to stochastic idea generator. The hallucinations, perhaps, tickle a deeper truth with the problem of induction: using past data to predict results has limits.